Passer en français
Start with Coach

Trust & data

What AI does in Coach, what it doesn’t, and where the data goes.

Six principles, each with how it is held in the product, then the answers about data: what, why, where, for how long.

In brief.

  • What Coach usesThe year, the subject, the interests chosen, and the work done in Coach.
  • What it doesn’t ask forAddress, phone number, photo.
  • What goes to the AI modelsThe learning profile (first name, year, interests, preferences) and what the student wrote. No e-mail address and no account identifier.
  • To deleteOne message through the contact form. The account and its history are deleted.
  • To ask a questionThe contact form. A member of the team replies.

Six questions we are asked, and how each is held.

  1. What role does AI have in learning?

    That of a support between two lessons: extending classroom work when a teacher cannot be beside every student. The content follows the Ministry of Education’s official programmes, published in the Bulletin officiel.

    How it’s held: Coach offers the chapters of the year’s curriculum.

  2. Why doesn’t Coach give the answer during an exercise?

    Because a student learns nothing from an answer handed over. The goal is for them to reach the solution themselves, with the help they need, then see the detailed solution to compare.

    How it’s held: during help on an exercise, and on a practice Brevet or Bac in assisted mode, the solution is not sent to the model; the reply must follow an imposed format (method steps, a similar example with other numbers) and an automatic check rejects a reply that would contain the solution, the statement or its numbers. It is a safeguard, not an absolute guarantee: the model can still solve an exercise in its own words. The Companion beside a lesson and the AI tutor outside an exercise have no such restriction.

  3. Where is the human judgement?

    With the pilot teachers, who observe in class what Coach does with their students. In the teacher tools in development, the AI will propose and the teacher will decide.

    How it’s held: for Teacher Assistant, a design commitment: no automatic action on a grade or a pathway without validation.

  4. How do you avoid dependence on the tool?

    By requiring success without help before moving on. On a given notion, the student should need Coach less and less.

    How it’s held: the exam is taken without the Companion, and the challenge comes after. Coach sends no notification to bring the student back.

  5. What if the AI is wrong?

    It can be. Generated exercises, problems, challenges and papers go through automatic checks before they are shown; practice Brevet and Bac papers are also solved blind by a second model. Corrections are written by the AI from the student’s reasoning. An error reported through the contact form is corrected, and we reply.

    How it’s held: a validation agent and deterministic checks on every piece of generated content; blind solving of exam papers; correction of reported cases.

  6. How do you design for minors?

    With rules: readable information, protective settings by default, no mechanism designed to retain.

    How it’s held: plain-language information; no advertising profiling; no pressure notifications.

What we will never use.

A product for minors must also refuse to manipulate. The mechanisms we have decided not to use:

  • To retain the student

    • Daily streaks not to break
    • Guilt messages (“Coach is waiting for you”)
    • Pressure notifications

  • To sell to parents

    • Countdowns and false scarcity
    • Friction at cancellation
    • Pre-ticked marketing consent
    • Fear sold to parents (“your child is falling behind”)

How exercises and papers are checked.

Lessons are reference content. Exercises, problems, challenges and exam papers are generated by the AI within the curriculum, then checked before they reach the student.

  1. Automatic checks before display

    A validation agent reads every piece of generated content, and deterministic checks verify its consistency with the curriculum and the student’s year.

  2. Blind solving of exam papers

    A practice Brevet or Bac is solved by a second model, without seeing the answer key, before it is offered.

  3. Correction of reported errors

    Errors reported through the contact form are corrected.

The path of a student’s question.

  1. Student

    One question, one attempt

    Year, subject, chapter, what the student wrote, and their learning profile: first name, interests, preferences.

  2. Models

    Generation

    The models receive this request and generate the reply. No e-mail address or account identifier is sent to them.

  3. AiCole

    Learning history

    Attempts, grades and feedback are kept with us, linked to the student’s account.

The answers, one per line.

What a parent or a head of school asks, in the order they ask it.

CollectedYear, subject, interests chosen at sign-up, work done in Coach, the account’s e-mail address.
Not collectedPostal address, phone number, photo, geolocation, data from other applications.
WhyTo adapt the lesson, the help and the correction to this student. Nothing else.
AI modelsReceive the learning profile (first name, year, interests, preferences) and what the student wrote. No e-mail address or account identifier.
HostingData hosted by Hetzner, in Germany. AI models hosted in France, on Microsoft Azure.
RetentionAccount data is kept according to the privacy policy, and deleted on request.
ProcessorsModel and hosting providers: Hetzner (data, Germany) and Microsoft Azure (AI models, France).
AdvertisingNone. No advertising profiling, no resale of data.
GAR, SSO, ENTGAR under evaluation. Sign-in by e-mail or Google account. No ENT integration for now.
SecurityArchitecture aligned with ISO 27001, not certified.

Your rights, and how to exercise them.

One message is enough. Reply within 30 days at most, often much less.

Send a request 

  • See an account’s data

    Full export on request

  • Correct

    From the account, or through the contact form

  • Delete

    Account and history, on request

  • Withdraw consent

    At any time, same effect as deletion

  • Complaint

    With the CNIL, without going through us

The regulatory frame, and where we stand.

  • GDPR

    Data reduced to what is necessary. Register, DPA and documentation on request.

  • CNIL, minors

    Information adapted to minors, no advertising profiling.

  • European AI Act · analysis in progress

    Coach is a support tool: it does not assess students for orientation purposes. The school layer planned for 2027 will be designed with the classification of assessment uses in mind.

  • Ministry framework for AI in education

    Educational purpose, critical thinking, data protection, professional judgement. The terms of classroom use (teacher supervision, year groups concerned) are defined with each school.

To go further.